Industrial Engineering

Dumb, Smart or Intelligent? What’s Really Relevant in the Factory

Dumb, Smart or Intelligent? What’s Really Relevant in the Factory
Industry Week July 2021
By Arno Koch
https://www.industryweek.com/operations/continuous-improvement/article/21170962/dumb-smart-or-intelligent-whats-really-relevant-in-the-factory

Machines used to be dumb. They didn’t make decisions, nor could they learn anything new. That is changing rapidly.

Smart

With the integration of software and sensors, a machine can make contact with its environment, and human “intelligence” can be introduced in the form of decision-making algorithms. This enables the machine to respond to its environment and make decisions.

A very simple pre-computer example is the loom that formed the basis of the later Toyota company: The machine could detect when a tread had broken and would stop the weaving process. Simple yet smart: with the machine detecting a problem and stopping the process, it no longer needed to be supervised permanently by a human.

Modern cars can distinguish the front from the back of other cars and determine their location and then calculate whether the other car is on a collision course or not. Based on built-in decision rules, the car chooses to give the driver a warning, take over the controls and change the route slightly or even make an emergency stop. Smart. But never smarter than the intelligence a human has put into the system.

Sensory awareness enables smart decisions 

Smart machines have sensory awareness of their environment and can make decisions based on decision rules integrated into the equipment. The better this works, the smarter the machine is. But the system itself does not learn; people do that for the system.

Intelligent

Machine learning. When a smart machine is able to change decision rules to improve its decisions, we speak of “machine learning.” The machine is able to learn and become better and better at its decisions.

Suppose the car from our example changes the interventions based upon what it learned from previous interventions: The machine ‘learns’ from previous experiences and gets better and better.

Machine learning is a great way to not only quickly improve the machine’s response, but also to make it responsive to changing conditions.

While the learning algorithm remains the same, the decision process may become more precise and adequate, but the machine will never ultimately outperform what the learning algorithm allows.

Artificial intelligence. It becomes exciting when the machine is also capable of improving the learning process itself. The machine could start looking for patterns on its own and look for other parameters that influence it. For example, it could discover that a sudden change in tire pressure at the rear (heavy load) affects braking and steering behavior and learn that “when that happens, I’d better intervene sooner and faster so that the car remains under control.”

In machine learning, the “learning rules” are static: “If this, then that.” In real intelligence, however, the learning rules consist of principles. The machine applies these principles to its reality and learns from the result of its response what the best decision would have been, and then formulates new decision rules from that.

Such self-learning systems are getting better and better, but for now we still need intelligent people. And as the 737 Max debacle has taught us, smart and intelligent systems are not a solution for fundamental system flaws.

Are “smart” and “intelligent” always a plus?

It is just like with people: There are super intelligent people who hardly do anything with their ability (they are not very smart) but there are also people with little intelligence, but who are smart in the sense they optimally convert what little they have in order to become better.

Ideally, the machine has enough intelligence on board to improve its decision rules and turn those decision rules into beneficial decisions, which then lead to better output.

Some Conclusions

1 . The labels “smart,” “intelligent” and “4.0” are, unfortunately, buzzwords mistakenly placed on all kinds of products. And though technology might be very helpful, don’t expect it to miraculously solve problems that should have been solved without the extra complexity.

  1. Adding investments and complexity to solve problems is rarely a good idea; simplification, standardization, stabilization, engaging people and reduction of non-value-adding activities are good ideas.
  2. Adding new technology can create value when it allows creating or achieving things such as precision or speed that couldn’t have been achieved without that technology.
  3. As long as machines are at best somewhat smart, people around the machine must be enabled to convert intelligence into smart decisions. And when such decision rules cannot be built into the machine, they have to be built into the process: That is the “A” in PDCA (Plan Do Check Act): Standardizing an improvement that follows from continuous improvement.

What to Do Now?

In every factory, choices must be made and priorities set. By applying further process improvements leading to autonomation—automation with a human touch—and/or simplification, revenue can be gained without capital investments, at same time laying a rock-solid foundation for upcoming new technologies,

Recommendation 1:

At first, deploy (new) technology only to create (more) value; to do things that bring more value to the customer, and which could not have done without this technology. Never use technology to organize problems; instead, work to eliminate them.

Recommendation 2:

Visualize losses (e.g. all that your customer is not willing to pay for) and solve their sources and root causes. Quit “stopgap measures” and “organizing” structural problems.

Recommendation 3:

Put your processes in order first; make them fast, flexible, reliable and effective using the craftsmanship and creativity of your crew. Only then start automating for quantitative or “additional feature to the product” reasons. Automation is not a solution for bad processes!

Recommendation 4:

Don’t make manufacturing an engineers’, management or—worse—suppliers’ domain. Empower all employees who work to keep inefficient processes running. They suffer the most from those processes and have the most inside knowledge about them, and so can best come up with ideas to change them.

Recommendation 5:

Put technology at the service of people and process. Let technology create value! However fascinating and even tempting some new technologies may be, at the end of the day the question is: Does it really help to achieve fundamental sustainable improvement, reduce risks and make the organization rock—solid and ready for our challenges? Is the new technology organizing a problem or really adding value?

Recommendation 6:

Keep things within the human scale: If you can’t explain it to your neighbor, it is probably too complex.